I don't know if that will change, but right now only the Voxtral Mini Transcribe V2 supports diarization and it's not open-weight.
The Voxtral Realtime model doesn't support diarization, but is open-weight.
You can test it yourself for free on https://console.mistral.ai/build/audio/speech-to-text
I tried it on an english-speaking podcast episode, and apart from identying one host as two different speakers (but only once for a few sentences at the start), the rest was flawless from what I could see
I think what he is saying is adding simple svg icons, not images.
For instance this one
https://www.svgrepo.com/svg/308550/earthquake-damage-earthqu... is 700 bytes, and it can most likely be simplified and compressed.
(you can also lazy load them so they don't delay the content)
Yes and the discussion here is about how that's not something that might be possible in the future, based on M1 and M2 support still being partial and M3 really experimental
Thank you for this, I updated some time ago but never really switched. Night and day difference !
The other thing I'm waiting for is search results ordered by date instead of relevance. When I'm searching for a picture in particular I know was taken 3 years ago, and search keywords to find it, it's impossible to find this specific photo because the ordering seem random
> "He explained that I was looking at a plan for a restored ecosystem. It was also a perfectly adapted money machine. There was a large area where the ancient elephants could graze, and this would be funded, in part, by carbon-offset payments from governments and corporations. The carbon value of a single elephant is about two million dollars, he told me. (An elephant increases biodiversity, in part, by spreading seeds in its dung and by crushing dense vegetation on forest floors, giving slow-growing trees the space to survive.) He added that the interesting educational opportunities and “sexiness factor” of Colossal’s creations would make its carbon credits “trade at a premium.”"
So it's a startup, valued at 10 billion?!
How exactly do they plan to make money?
Seriously, could anything be more 21st-century? Resurrecting extinct animal species (ones that supposedly went extinct naturally, mind you, not because of humans – what's the point then?) just to reintroduce them into parks and sell carbon credits.
The real prize is the technology and techniques to do this sort of stuff. CRISPR is a fascinating technology that we're just now seeing the benefits of[0]
From the article, it looks like they have multiple teams working on multipe animals at the same time. But the dodo team is going slower than the mammoth and the wolf :
> Keyte added that her team was still a long way from bringing back the dodo. For one thing, the methods for growing and manipulating the embryonic precursors of avian sperm and eggs in a lab setting have been developed for only two birds: the chicken and, recently, the goose. Keyte said, “It’s been almost twenty years since culture conditions for the chicken were established, and those culture conditions have not worked for other bird species, even ones that are really closely related, like quail.” She added that, despite the dearth of related research, her team was getting better at growing the sperm-and-egg precursors in birds: “We’ve gotten to the point where we feel like we can start doing some migration assays”—a technique for studying how the cells in an early embryo begin to differentiate. Once the researchers got the basic method for growing bird cells down, they could use the technology not just to develop a dodo but also to help replenish populations of endangered birds. The team had already identified some species that could use the help.
reply